NK cells in peripheral blood carry trogocytosed tumor antigens from solid cancer cells

The innate immune lymphocyte lineage natural killer (NK) cell infiltrates tumor environment where it can recognize and eliminate tumor cells. NK cell tumor infiltration is linked to patient prognosis. However, it is unknown if some of these antitumor NK cells leave the tumor environment. In blood-borne cancers, NK cells that have interacted with leukemic cells are recognized by the co-expression of two CD45 isoforms (CD45RARO cells) and/or the plasma membrane presence of tumor antigens (Ag), which NK cells acquire by trogocytosis. We evaluated solid tumor Ag uptake by trogocytosis on NK cells by performing co-cultures in vitro. We analyzed NK population subsets by unsupervised dimensional reduction techniques in blood samples from breast tumor (BC) patients and healthy donors (HD). We confirmed that NK cells perform trogocytosis from solid cancer cells in vitro. The extent of trogocytosis depends on the target cell and the antigen, but not on the amount of Ag expressed by the target cell or the sensitivity to NK cell killing. We identified by FlowSOM (Self-Organizing Maps) several NK cell clusters differentially abundant between BC patients and HD, including anti-tumor NK subsets with phenotype CD45RARO+CD107a+. These analyses showed that bona-fide NK cells that have degranulated were increased in patients and, additionally, these NK cells exhibit trogocytosis of solid tumor Ag on their surface. However, the frequency of NK cells that have trogocytosed is very low and much lower than that found in hematological cancer patients, suggesting that the number of NK cells that exit the tumor environment is scarce. To our knowledge, this is the first report describing the presence of solid tumor markers on circulating NK subsets from breast tumor patients. This NK cell immune profiling could lead to generate novel strategies to complement established therapies for BC patients or to the use of peripheral blood NK cells in the theranostic of solid cancer patients after treatment.


Introduction
Natural killer (NK) cells are a subset of lymphoid cells and part of the innate immune compartment. As blood-circulating cells with cytotoxic activity, NK cells screen for damaged or stressed cells, and they are readily able to kill virus-infected or transformed tumor cells, contributing to immune surveillance (1). These cells are mainly classified as CD56+CD3-innate lymphoid cells, but they constitute a heterogeneous population comprising NK cell subsets with different cytotoxic potential. Based on CD56 and CD16 surface expression, they subdivide into CD56 + CD16 high blood-circulating NK cells, with stronger cytotoxic activity after target cell recognition, and CD56 high CD16 low cells cytokine-producing NK cells with poor cytolytic activity, mostly present in secondary lymphoid tissues (2). However, recent studies have challenged this classical view, and high dimensionality, single-cell proteomic analysis have revealed a striking NK cell phenotype diversity, which might be influenced by both genetic differences between individual humans and environmental conditions (3). Considering that NK cell function is tightly modulated by the expression of several inhibitory and activating receptors, the determination of NK cell subsets repertoire and their contribution to physiological processes could be of paramount importance for generation of NK cell-based therapies against malignant diseases (4,5).
The anti-tumor properties of NK cells have been previously discussed (6). NK cells play an essential role in tumor clearance by recognizing and killing abnormal tumor cells without the need of prior activation. After recognition of target cells, different target cell-derived proteins can be acquired in NK cell membrane surface in a cell-to-cell contact-dependent manner, a process called trogocytosis (7)(8)(9)(10)(11). Trogocytosis involves an intercellular transfer of membrane patches, and it has been shown to occur in different immune cell types, albeit the physiological relevance of this process is not fully understood (9,(11)(12)(13). This transference of functional proteins to cell surface could modulate NK function in vitro and in vivo (14-20). Trogocytosis is receiving high interest from the clinic for this possibility to modulate NK cell (18,20) or CAR T cell function (21,22).
We have reported the identification of NK cell populations with anti-tumor activity in hematological cancer patients (23)(24)(25)(26). The highly activated CD56+CD16 high NK cells found in these patients exhibit the expression of activation markers, such as NKp46 and NKG2D, and low expression levels of inhibitory markers, such as NKG2A and CD94. Interestingly, these NK cells also present non-NK, tumor cell-derived antigens on their surface, which can be an indicative of trogocytosis during cell killing (23)(24)(25)(26). We found that this subset of anti-tumor NK cells is also characterized by degranulation and co-expression of both CD45RO and CD45RA (CD45RARO cells) (23)(24)(25)(26). Further high-dimensionality, multiparametric flow cytometry and unsupervised analyses in multiple hematological tumor patients showed that NK subsets presenting CD45RARO phenotype and evident tumor-antigen derived trogocytosis responds directly to the oncologic status of patients, which suggest that the frequency of the function of these NK subsets depend of the presence of targets (23,24).
NK cells infiltrate solid cancers, as well as tumor-infiltrated lymph nodes and metastases (27)(28)(29). NK cell infiltration of most solid tumor is rather sparse and depends on tumor localization and the nature of the cancer (30). To our knowledge, detection of antitumor NK cell subsets expressing trogocytosed tumor-derived markers in peripheral blood of solid tumor patients have not been reported. Here we investigated whether NK cells could acquire solid tumor antigens by trogocytosis in vitro and in vivo. Additionally, we studied the presence of anti-tumor, bloodcirculating, NK cells subsets exhibiting these solid tumor antigens in breast cancer patients by multiparametric flow cytometry and high-dimensionality unsupervised analyses.

Ethical statement
The use of human specimens for scientific purposes was approved by the French National Ethics Committee. All methods were carried out in accordance with the approved guidelines and regulations of this committee. Written informed consent was obtained from each patient or donor prior to collection.

Breast tumors patients
Data and samples from patients were collected at the Institute for Cancerology of Montpellier (ICM), France, after patient's written consent and following French regulations. Patients were enrolled in the ICM-BDD 2017/37 (ID-RCB: 2017-A01940-53) clinical program approved by the "Comiteś de Protection des Personnes Sud-Ouest et Outre-Mer III" with the reference 2017/ 45. Blood samples were collected at diagnosis. Peripheral blood mononuclear cells (PBMCs) were obtained by Ficoll® gradient and stored frozen in liquid nitrogen until use. Patients' status is described in Table 1.

Healthy donor
HD samples were obtained from written informed donors, collected by clinicians of the CHU Montpellier and collected and processed as the patient's samples.

Cell lines
Breast cancer cell lines BT-20, SKBR3 and MDA-MB-468, pancreatic adenocarcinoma cell line LNCaP, colorectal adenocarcinoma cell line HCT116, and the Epstein-Barr Virus (EBV)-transformed lymphoblastoid B cell line PLH were grown in RPMI 1640 media (Gibco) supplemented with 10% fetal bovine serum (FBS). Cells were used for experiments at confluency of 80%. Cell line identity was confirmed by flow cytometry when possible, and cells were regularly tested for mycoplasma.

UCBMC purification
Umbilical cord blood (UCB) units obtained from healthy donors from CHU Montpellier. UCB mononuclear cells

Enrichment, activation and expansion of human NK cells
Expanded NK (eNK) cells were obtained as previously described (31)(32)(33)(34). Briefly, UCBMC were depleted of T cells by using EasySep ™ CD3 Positive Selection Kit II (STEMCELL Technologies). Cells were cultured in the presence of g-irradiated PLH cells at ratio NK-to-accessory cell of 1:1 in RPMI 10% FBS media supplemented with human IL-2 (100UI/mL, Peprotech) and human IL-15 (5 ng/mL, Miltenyi Biotec) for 14-to-21 days. Once every 3 days, cells were counted and fresh culture media with FBS, IL-2 and IL-15 was added to the culture, along with additional girradiated PLH cells. Purity of human CD3-CD56+ eNK cells at the end of the culture was always ≥ 90%. For experiments to determine measuring membrane dye transfer, cells were for surface markers with the antibodies CD56-V450 and CD3-FITC (all from BD). Cell viability was determined using 7-AAD exclusion (Miltenyi Biotec). Staining was performed in FACS buffer at 4°C for 25-30 min, and then cells were washed three times before FACS acquisition and analysis in Gallios flow cytometer instrument (Beckman Coulter). After exclusion of doublets and dead cells, tumor marker detection on CD3-CD56+CD335+ NK cells was evaluated by analyzing FCS files using FlowJo software v10.6.1 (Tree Star Inc.).

Fluorescence microscopy
Breast cancer cell line SKBR3 cells were seeded in 24-well flatbottom plates (200,000 cells/mL in culture medium) on coverslips previously coated with poly-D-lysine (Sigma), and incubated overnight at 37°C. Trogocytosis in vitro of tumor markers was attained by co-culturing eNK cells (acceptor cells) with SKBR3 (donor cells) in effector-to-target ratio of 1:1, leaving some coverslips with eNK alone. Cells were incubated overnight at 37°C and after media removal and wash with PBS, attached cells were fixed using 4% paraformaldehyde diluted in PBS and blocked with 5% FBS diluted in PBS. Cells were then incubated with CD56-AlexaFluor ™ 488 and CD326-PE antibodies (all from BD Biosciences) diluted in blocking solution for 2 h at room temperature, protected from light. Some coverslips with eNK cells were stained with CD45-PE antibody (Beckman) diluted in blocking solution. Hoescht 33342 was used for nucleus staining. After several washing steps, coverslips were mounted on microscope glass slides using Prolong Gold mounting media (ThermoFisher). Cell samples were visualized using a Leica SP5 fluorescence microscope (Carl Zeiss, Germany) and images analyzed using the Las X Life Science software (Carl Zeiss).

High dimensional reduction analysis
To generate tSNE or UMAP embedding, a pre-gated NK cell population from each sample with the same number of cells per patient and timepoint was selected using FlowJo Downsample plugin (v3.1.0) and merged before uploading in the Cytobank cloud-based platform (Cytobank, Inc.). High-dimensional single-cell data dimensionality reduction was performed by viSNE, which is based upon the t-Distributed Stochastic Neighbor Embedding (t-SNE) implementation of Barnes-Hut (35). viSNE was used to visualize FACS data as 2D t-SNE maps, using the following parameters: Desired Total Events (Equal sub-sampling): 50.000; Channels: selected all 16 surface markers; Compensation: "File-Internal Compensation"; Iterations: 1000; Seed; "Random"; Theta: 0.5. FlowSOM was used with default settings unless otherwise noted. FlowSOM uses Self-Organizing Maps (SOMs), based on marker expression phenotype, to assign all individual cells into clusters and metaclusters (that is, group of clusters) (36). FlowSOM was performed with the following parameters: Event Sampling Method: "Equal"; Desired events per file: "2.736"; Total events actually sampled: "27.160"; SOM Creation: "Create a new SOM"; Clustering Method: "Hierarchical Consensus"; Number of metaclusters: "12"; Number clusters: "256"; Iterations: "100"; Seed: "4567". CITRUS (cluster identification, characterization, and regression) is an algorithm designed for the discovery of statistically significant stratifying biological signatures within single cell datasets containing numerous samples across associated conditions or correlated with clinical phenotype of interest (e.g. responders versus non-responders) (37). The output is a network topology of cell subpopulations divided in sub-clusters that represents a hierarchical stratification of the original sample. Median expression levels of functional markers measured across each population can drive the differentiation between phenotypes. CITRUS was performed using the Significance Analysis of Microarrays (SAM) correlative association model (Benjamin-Hochberg-corrected P value, false discovery rate (FDR) < 0.01), with the following parameters: Clustering channels: "selected all surface markers except CD107a and tumor markers"; Compensation: "File-Internal Compensation"; Statistic channels: "CD107a and tumor markers"; Association Models: "Significance Analysis of Microarrays (SAM) -Correlative"; Cluster Characterization: "Medians"; Event sampling: "Equal"; Event sampled per file: "2000"; Minimum cluster size (%): "1"; Cross Validation Folds: "5"; False Discovey Rate (%): "1". Identification of NK cell subsets between the group "Healthy Donors" and "Breast Cancer patients" was performed by comparing the relative expression of CD107a and tumor markers of the specified FlowSOM metacluster.

Statistical analysis
Experimental figures and statistical analysis were performed using GraphPad Prism (v8.0). All statistical values are presented as * p<0.05; ** p<0.01; *** p<0.001 and **** p<0.0001. Mean values are expressed as mean plus or minus the standard error of the mean (SEM).

NK cells perform trogocytosis in vitro on solid tumor cells
The breast cancer cell lines BT-20 and SKBR3, the human prostate adenocarcinoma cell line LNCaP and the human colon cancer cell line HCT116 lack expression of CD3 (T cell marker) and CD56 (NK cell marker) and show very low expression of CD335 (NK cell marker), also known as NKp46/NCR1 ( Figure S1). In contrast, they do express human epidermal growth factor receptor 2 (HER2) and Epithelial cell adhesion molecule (EpCAM or CD326).
As expected, eNK do not express HER2 or prostate-specific membrane antigen (PSMA) ( Figure 1A). When eNK were incubated with BT-20 and SKBR3 cells, they gained HER2 expression ( Figure 1A), although the levels were 10 times lower that those found in target cells ( Figures S1, 1B). eNK also gained PSMA when incubated with LNCaP cells that largely express this Ag ( Figures 1A, C). Again, the expression of PSMA was 10 times lower for eNK than for LNCaP ( Figure 1C). Optimal trogocytosis on all cell lines was observed at 3:1 E:T ratio ( Figures 1D, S3). Of note, eNK encountering BT-20 cells acquire less HER2 expression than those encountering SKBR3 ( Figure 1A, S3), although the tumor cell sensitivity to NK cytotoxicity was similar ( Figure S2).
After confirming that NK cells can capture tumor antigens from solid cancer cells, we wanted to elucidate whether this gaining of expression was caused by trogocytosis. For this objective, as trogocytosis is a very rapid process, we restricted our co-cultures down to 2 h of incubation (38,39). There is a lack of specific inhibitors for trogocytosis, but it has been described its interference by disruption of actin polymerization, inhibition of kinases (such as Src-kinase and Syk-kinase) and low temperature (4°C) (40,41). Before performing the co-culture with SKBR3 tumor cells, we pretreated eNK cells with cytochalasin-D (CytD) for inhibition of actin recruitment, or incubated the cells in the presence of PP2 for inhibition of Src-tyrosine kinase, which resulted in a decrease of HER2 acquisition by NK cells ( Figure S4A). Moreover, we carried out these co-cultures at 4°C which completely reduced tumor marker acquisition ( Figure S4B). We complemented these observations by using MDA-MB-468 as breast cancer donor tumor cells, which are negative for HER2 in comparison to SKBR3 ( Figure S4C). Consistent with the idea that HER2 is acquired from donor cells via trogocytosis, NK cells co-cultured with MDA-MB-468 cells at different ratios did not increase HER2 expression after incubation ( Figure S4D). However, this result would not be due to the absence of trogocytosis from this cell line, as we confirmed by membrane dye transfer experiments, in which NK cells were co-cultured with SKBR3 or MDA-MB-468 breast tumor cells previously labeled with the lipid intercalant dye DiD ( Figure S4E). Even only after 2 h of co-culture, NK cells became strongly positive for this dye, and pre-treatments with CytD and PP2 along with 4°C incubation blocked the acquisition of donor membrane lipids from NK cells ( Figure S4E). Considering that transfer of proteins via trogocytosis goes together with transfer of membrane lipids, altogether these results suggest that HER2 and other solid tumor markers are acquired by NK cells via trogocytosis.
To further confirm that eNK cells captured tumor cellexpressed receptors by trogocytosis, we analyzed the interaction between solid tumor cells (donor cells) and eNK cells during co-culture by the approach of fluorescence microscopy. As expected, we observed that eNK cells were positive for CD56 and CD45, and negative for EpCAM (Figures 2A, B), while solid tumor cells SKBR3 (donor cells) expressed EpCAM (CD326, Figure 2C). eNK cocultured with SKBR3 cells interacted with them by forming immune synapses and performing cytotoxicity. After trogocytosis in vitro, eNK acquired surface expression of EpCAM. (Figure 2D), which was absent in NK cells incubated alone. This surface expression was still evident after several washing steps ( Figure 2D).
We next investigated if different Ags were differently uploaded from the same targets. eNK efficiently gained EpCAM, but much less HER2, from HCT116 or LNCaP cells (Figure 3), whereas these cells express substantial amounts of both Ags ( Figure S1). But this difference does not depend on HER2, because both HER2 and EpCAM are efficiently uploaded from SKBR3 cells (Figure 3). In summary, the efficiency of trogocytosis is variable regarding to the target cell, the efficiency of killing and the type of Ag.

Identification of trogocytosis ex vivo
We used a cohort of patients operated for a breast tumor (8 invasive breast cancers, one ductal in situ carcinoma, one atypical ductal hyperplasia) to analyze several phenotypic markers on The percentage of CD7 + cells, which mainly include T and NK cells tended to decrease in patients; and in fact, the percentage of CD56 + cells decreased (Figures 4A, B). However, the CD56 + /CD16 + cells, which represents the mature NK cells, remained stable ( Figure 4C). The CD56 + cell subset that decreased were CD3 + and should represent NK T cells ( Figure 4D). In contrast, the CD56 + /CD3population, which represents NK cells, remained stable.
In hematological cancers, the main antitumor NK cell population is recognized by the expression of CD45RO (CD45RO cells), generally together with CD45RA (CD45RARO cells (23)(24)(25)(26);. We observed a small, but significant, increase of this CD45RARO population in peripheral blood NK cells of patients ( Figure 4E). This was associated with a decrease in cells expressing low CD45RA levels (CD45RA dim ) and CD45RO (CD45RA dim RO cells). On the other side, the CD45RO + NK subset frequency was found similar between BC patients and HD ( Figure 4E).
We next evaluated degranulation, i.e. CD107a + cells, in the lymphoid, i.e. CD7 + , compartment and observed an increase in patients ( Figure S6A). Exclusion of CD14 + /CD19 + cells did not change our observation ( Figure S6B). The differences were not statistically different when we focused on receptor-negative patients ( Figure S6C) and hence, the increase mainly relied in receptorpositive patients (Figures S6D, E). Analysis of bona fide CD7 + CD56 + CD3 -NK cells (42) confirmed the higher degranulation of NK cells in BC patients ( Figure 4F).
We analyzed trogocytosis of 4 endothelial/tumor markers, which are not expressed by NK cells. These were: HER2 (human epidermal growth factor receptor 2, which is overexpressed in certain patients), CD326 (EpCAM), Mucin 1 cell surface associated (MUC1) and PSMA. The membrane localization of all these proteins makes them candidates to be trogocytosed and exposed in NK cell plasma membrane. Patients showed higher trogocytosis of all markers except EpCAM ( Figure 5A), suggesting that this Ag is not well uptaken by NK cells in vivo. However, differences were not statistically significant in the total NK cell population ( Figure 5B), nor in the cells that had degranulated ( Figure 5C). Notably, when we focused the analysis on receptorpositive patients, the frequency of total CD7+CD56+ NK cells exhibiting the tumor markers MUC1 and HER2 appeared to be increased ( Figure 5D). The percentage of cells that have performed trogocytosis on the epithelial markers was significantly higher in CD107a + NK cells than in the bulk of NK cells, principally in patient's samples ( Figure S6F). We also observed this increase in CD45RARO cells regarding the bulk of NK cells ( Figure S6G). However, we did not observe significant differences between HD NK CD45RARO+ and BC NK CD45RARO+ populations, nor between CD45RARO+CD107a + NK cells from both groups ( Figure S6H). This indicates that CD45RARO and CD107a + cells are the NK cell populations that are interacting with target cells and recovering antigens by trogocytosis. Although there were not significant differences between receptor-negative BC patients and HDs, some of the tumor markers analyzed were found upregulated on CD7+CD56+ NK cells from receptor-positive BC patients when compared with HD ( Figure 5D). The fact that HDs show a similar pattern suggests that NK cells can interact and probably kill endothelial cells, which could be stress or damaged cells.

Differential trogocytosed-receptor expression pattern in NK cell subsets between BC patients and healthy donors by viSNE
In order to further characterize the expression of these tumor markers present on NK cell surface due to trogocytosis, we used the high-dimensional reduction algorithm viSNE (for visualization of t-Distributed Stochastic Neighbor Embedding or t-SNE) (35,43). By viSNE, we generated unsupervised 2D t-SNE maps showing the expression of 12 surface markers on NK cells from BC patients and healthy donors ( Figure S7). The resulting t-SNE maps exhibit several spatial regions with differences in marker expression, suggesting that NK cell subsets with different abundance and expression of those markers were present.
To evaluate the differences in NK cell subset distribution between BC patients and healthy donors, we applied the FlowSOM method which uses Self-Organizing Maps (SOMs) to cluster together cell events based on clustering channels (markers) and assign them to metaclusters, grouping them into distinct populations by an unsupervised approach (36). By this method we identified twelve metaclusters, of which two, i.e. 8 and 12, were found to be differentially abundant between BC patients and healthy donors ( Figures 6A, B, Table S1). Both mainly consisted in NK cells with phenotype CD56 + CD16 high CD45RA + CD107a + and tumor markers expression (Table S1), Metacluster-8 was found enriched in BC patients in comparison with healthy controls and contained CD45RO + cells (10.7% versus 6.9% of control, p<0.05) ( Figure 6B). This difference was even more evident when we compared receptorpositive patients with healthy controls (12.4% versus 6.9% of control, p<0.01) ( Figure S8A, Table S1). On the other side, metacluster-12 was found to be decreased in BC patients compared to healthy donors (15.5% versus 19.5% of control, p<0.05). This difference was higher when controls were compared with receptor-negative patients (13.2% versus 19.5% of control). In contrast to metacluster-8, this metacluster-12 mostly comprised CD45RO -NK cells (Table S1). When we analyzed the degranulation on NK cells, we found that CD107a expression levels on Metacluster-7, Metacluster-8 and Metacluster-12 were significantly higher in BC patients compared to controls ( Figure 6C). CD107a+ was also found increased in NK cells from Metacluster-10 when comparing receptor-positive BC patients and healthy donors ( Figure S8B). Thus, considering that NK cells having a CD45RARO/CD107a+ phenotype previously exhibited higher degree of presumed trogocytosed proteins, we analyzed the expression levels of these solid tumor markers on these metaclusters. Interestingly, among all FlowSOM-identified metaclusters, we found that NK cells contained in Metacluster-8 showed the higher levels of tumor markers and, compared to healthy controls, BC patients exhibited significantly higher levels for the tumor markers CD326 and PSMA ( Figure 6C). Moreover, HER2 was found particularly increased on NK cells from Metacluster-8 of receptor-positive BC patients in comparison to healthy donors ( Figure S8B). Lastly, the expression of the immune checkpoint receptor programmed cell death-1 protein (PD1) was also analyzed on these NK cells, and when we compare its expression between healthy control and BC samples (total or receptor-positive patients), there was not significative difference between both groups (Figures 6C, S8B). Altogether, these results suggest that bona-fide NK cells with an activated phenotype (CD107 + CD45RARO) are found increased on BC patients and, additionally, these NK cells exhibit high degree of potentially trogocytosed tumor markers on their surface. To complement these observations with FlowSOM analyses, we further examinate the cell activation and trogocytosis profile of BC patients and controls, focusing on NK cells contained in Metacluster-8 and using CITRUS (37). This algorithm allows the identification of stratifying sub-populations in multidimensional flow cytometry datasets, and can be used to distinguish single-cell signatures that might be associated with clinical outcomes (44). clusters, of which four were found to be statistically associated to the BC patient's group, in accord to their CD107a and tumor marker expression ( Figures 7A-D). By associating this CITRUS map with maps showing surface phenotype markers intensities ( Figure 7E), we could identify these clusters specifically as subsets of CD56+ NK cells expressing high levels of CD16, CD45RA, NKp46 and intermediate levels of CD45RO, with high expression of CD326, PSMA, HER2 and MUC1 (Figure 7).

Discussion
Individual humans have their own NK cell repertoire, which changes during development and is different in diverse tissues. In addition, multifactorial environmental events affect NK cells and generate new subsets (45) and make challenging to identify NK populations associated to a specific disease and shared by multiple patients. Conventional NK cells, which should correlate with the B C A FIGURE 6 Identification of NK cell clusters with acquired tumor markers by FlowSOM. FACS samples from 5 healthy donors and 10 BC patients were concatenated and randomly subsampled into 100.000 total events which were analyzed by t-SNE and displayed using viSNE. (A) FlowSOM-identified twelve metaclusters visualized in the concatenated file. To the right, metacluster-derived gating applied to concatenated HD and BC patient samples. Differences in color on the viSNE map correspond to cell abundancy density. (B) Comparison of relative frequency of each FlowSOM metacluster between HD and BC patients. (C) Median fluorescence intensity (MFI) of CD107a, PD1 and trogocytosed tumor markers expressed on CD7+CD56+ NK cells present on selected metaclusters (7, 8 and 12). Graph represent box and whiskers (Min to Max), and statistical significance between HD (n = 5), BC (n = 10) was determined by two-way ANOVA; * p ≤ 0.05; ** p ≤ 0.01. ns, not significant. population that we have studied, have a short half-life (5,46). Hence, the amount of NK cells that has infiltrated the tumor and interacted with targets, then performed trogocytosis and came back to the blood stream should be scarce. In this current work, we have found a very low number of NK cells carrying tumor Ags, showing that their numbers are very low or, alternatively, to find and identify them is extremely difficult. Moreover, we observed that NK cells from HD also carry some Ags typical of endothelial cells that we have used as trogocytosis markers. Probably, these NK cells have interacted with "stressed" endothelial cells. In case of breast cancer patients, the amount of "stressed" endothelial cells should be larger, because tumor cells are supposed to be stressed and recognized by NK cells, increasing the portion of NK cells that carry endothelial markers. However, most of the NK cells that have interacted with CITRUS identifies NK cell clusters with differential expression of tumor markers on BC patients. Cluster identification, characterization, and regression (CITRUS) algorithm identifies cell subsets (clusters) with significantly differential marker expression between BC patients and HD. tumor cells should be in the tumor microenvironment as shown previously (47). Here we show that effectively their frequency in periphery is very low, and to find differences between HD and patients we needed to unveil new populations (Figures 6, 7). Of note, NK cells eliminate senescent cells, which are considered "stressed", favoring the clearing of this population and keeping an adequate cell population (48). In support of this hypothesis, NK cells carrying endothelial markers have predominantly degranulated, i.e. they are CD107 + . This observation is consistent with previous reports describing higher degranulation and expression of activation markers on NK cells that acquired tumor cell markers by trogocytosis, proposing the surface expression of trogocytosed tumor markers as indicative of contact between these NK cells and marker-positive cancer cells (47,49). Our observations from in vitro experiments strongly imply that transference of HER2, EpCAM and other solid tumor cell markers on NK cells occurs via trogocytosis, proposing this process as the mechanism of capture of epithelial and tumor markers by NK cells under physiological and pathological conditions, as reported in tumor biopsies from BC patients (47). However, it is not possible to definitively conclude that HER2 and the rest of tumor markers found on circulating NK cells from BC patients would have been transferred via trogocytosis. These membrane proteins are normally absent and not expressed by NK cells, and our unsupervised multiparametric analyses found specific subsets of NK cells with high surface levels of these epithelial and solid tumor markers present in blood samples of BC patients; therefore, this correlation should not be completely ruled out, this point remaining pending of further elucidation.
In our study we focused on bona fide NK cells (CD7 + CD56 + CD3 -) from patient's peripheral blood. It has been reported that tumor-infiltrating monocytes and NK cells that can be found in breast tumors could show high HER2-trogocytosis, mainly when patients are treated with mAbs systemic treatment, which facilitate NK cell recruitment and activation (47). These observations make us suggest that the amount of Ag capture by trogocytosis on tumorinfiltrating NK cells could be higher and more measurable than the surface levels detected on circulating NKs. However, analysis of infiltrated, tissue-specific human NK populations could prove more challenging and complex (45). For this reason, we did not investigate tissue-specific NK cells that would require biopsies. Hence, we have not probably identified all disease-associated NK populations, but only those that could exit the tumor environment. In addition, we have used a defined panel of NK-associated markers, whereas NK cells can express hundreds of them (50). In summary, other anti-tumor NK populations probably exist and perform degranulation and trogocytosis.
The identification of subsets of circulating NK cells, consisting of activated cells with evident degranulation carrying epithelial and tumor cell markers, raises the issue of the relevance that holds this tumor-trogocytosed marker harboring on NKs cellular and immune functions. Several reports in mice observed that MHC-I acquisition from target cells decreases acceptor NK cell immune recognition and cytotoxic functions (51)(52)(53)(54). Correspondingly, it was reported that human NK cells that capture HLA-G (a non-classical MHC-I immunosuppressive molecule) from melanoma solid tumor cells not only massively reduce their proliferation and their cytotoxicity, but also are able to suppress the cytotoxic function of other bystander NK cells expressing the HLA-G ligand and inhibitory receptor ILT2 (14). Similarly, trogocytosis of CD9 molecules from ovarian carcinoma cells to NK cells renders them less cytotoxic and poorer producers of anti-tumor cytokines, consistent with the identification of a CD9-positive NK cell subset in tubo-ovarian carcinoma samples which presence correlates with tumor progression (55). In line with this, a recent report showed that murine NK cells can acquire by trogocytosis the immune checkpoint inhibitor programmed cell death protein 1 (PD1) from leukemia cells both in vitro and in vivo, suppressing antitumor NK cell immunity (20).
However, this might not be the case for tumor receptors like HER2, because similarly to our present results, it has been observed that NK cells co-cultured with trastuzumab-opsonized HER2+ breast cancer cells can acquire HER2 receptor via trogocytosis and exhibit higher expression of CD107a than non-HER2trogocytosed NK cells (47). In addition, NK cells are capable to obtain the tyrosine kinase receptor TYRO3 from leukemia cells in vitro and in vivo, displaying higher levels of activation markers, enhanced cytotoxicity and interferon-g secretion (49), thus providing opposing evidence that tumor receptor trogocytosis on NK cells could also translate into gain of anti-tumor activity and effector function (56). This study was focused on the identification of NK cell subsets harboring trogocytosed markers from breast cancer cells, and even if experiments to elucidate the functional outcome of the acquisition of these markers by trogocytosis are pending to perform, we observed high levels of CD107a and low levels of PD1 expression on these NK cells from BC patients, which considering the mentioned literature is consisting with high antitumor function of these NK cells.
We have found also that degranulation and trogocytosis is higher in patients that expressed estrogen and/or progesterone receptors. This could be related to the main historical view that triple-negative breast cancer (TNBC) is a cold tumor (57). Hence, it is expected that NK cell infiltration and interaction with tumor target cells is poor in these patients. This should explain the lower level of NK cells that have degranulated and performed trogocytosis in peripheral blood from these patients in our study. However, a larger cohort of patients including both estrogen/progesterone receptor-positive and -negative would be necessary to conclude with greater certain this point.
We expanded our observations by performing dimensionality reduction analysis with viSNE, and we complemented it by using FlowSOM and CITRUS algorithms to further determine NK cell subsets carrying tumor cell antigens. Our unsupervised analyses identified clusters over-represented in BC patients containing tumor markers-expressing NK cells, i.e. metacluster 8. This cluster consisted mostly of CD45RARO+ cells, with high level of degranulation that acquired surface expression of tumor markers, most probably by trogocytosis. These cells also exhibited relative expression levels of PD1, although we found no significative difference for this marker between BC and HD group (p = 0.61). Complementing this information, we have observed a very low PD1 levels (from 1% to 5% positive tumor cells) in our cell lines. This could suggest that PD1 is probably from NK cell origin in our settings, in accord with previous reports (58). Conversely, we cannot definitively rule out the likeliness that PD1 expression detected on NK cells derives from cancer cells, considering that PD1 expression on tumors derived from BC patients included in this study was not determined, and it has been already described that in different experimental settings NK cells can gain also PD1 by trogocytosis (20). Therefore, one possibility is that PD1 expression could increase on NK cells once they have acquired cancer cell markers. This could be due to NK cell activation after recognition of target cells, which have been reported on both circulating and tumor-infiltrating NK cells from several types of solid tumors (59-62). We favor this hypothesis, but we cannot exclude that PD1+ NK cells are more prone to acquire cancer markers. The physiological relevance could be important, because PD1+ NK cells with cancer antigens may participate in maintaining an immunosuppressive state and affect some immunotherapy treatments. Thus, trogocytosis of these types of receptors can show strong immunomodulatory capacities on NK cell immune function and capabilities.
Finally, our in vitro approach shows that different Ags are differently trogocytosed from different target cells and this is independent of the sensitivity to NK cells. NK cells extract for example HER2 and EpCAM from SK-BR3 or LNCaP cells; but almost exclusively EpCAM from HCT116 or BT20 cells. Meanwhile, the sensitivity of all these cell lines to NK cells was comparable. Hence, which molecules are trogocytosed depend on the donor cell and the nature of the molecule. The understanding of the precise mechanism of trogocytosis will be essential to unveil the reasons of the disparity between Ags and donor cells.

Data availability statement
The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding author.

Ethics statement
The studies involving human participants were reviewed and approved by ICM-BDD 2017/37 (ID-RCB: 2017-A01940-53) clinical program approved by the "Comiteś de Protection des Personnes Sud-Ouest et Outre-Mer III". The patients/participants provided their written informed consent to participate in this study.

Author contributions
WJ provided essential material and follow patient's status. GG, M-LD and MC performed the experiments. MC-M and MV performed study design and wrote the article. All authors contributed to the article and approved the submitted version.